Modeling Characteristics of Location from User Photos

Abstract

In the past decade, location-based services have grown through geo-tagging and place-tagging. Proliferation of GPS-enabled mobile devices further enabled exponential growth in geo-tagged user content. On the other hand, location-based applications harness the abundance of geo-tagged content to further improve user experience and more relevant localized content. We show in this paper that geo-tagged content can vary significantly based on whether they are captured by a local versus a tourist to the location. Using photos shared by online users, we also show how we can learn unique characteristics about a given location. We finally discuss an effective metric to rank the most representative photos for a given location by combining visual contents and their social engagement potential.